Author: wsm007 | Date: 2026-03-05 Event: Morgan Stanley Technology, Media & Telecom Conference 2026 Speakers: Adam Foroughi (CEO), Matt Stumpf (CFO) Format: Fireside chat with Matt Cost (MS U.S. Internet) Stock reaction: +10.0% to $482.81 on the day Source: Full Seeking Alpha transcript
Foroughi delivered the most compelling articulation of AppLovin's moat and growth path I've seen — the 50x creative gap quantifies the e-commerce unlock, the 1.3% → 5%+ conversion rate math frames a potential 4x+ in gross platform value, and the "recommendation systems will evolve like LLMs" framing is the 5-year thesis in a sentence.
This was the single most important disclosure in the fireside. I missed this entirely from the news summaries.
Foroughi quantified the creative differential: gaming developers run 50,000+ ads per campaign. E-commerce companies run 1,000 at best. That's a 50x gap. The model is built to test ads programmatically and find conversion rate expansion through creative diversity. With 50x fewer ads, e-commerce advertisers structurally can't exploit the platform's full capability.
The fix: Multi-agent generative AI for creative production — static images (piloting now) and video ("1-page animated GIF format," also piloting, full video model "on the way"). Foroughi: "Some point in the next 4 months, we will have both these types of problems solved — regenerative AI-based creative." That's a hard commitment to H1 2026.
Why this changes my model: At Q4 earnings I flagged the 43% qualified-lead breakage rate as a red flag, attributed to "creative format gap." Now I understand the mechanism: e-commerce advertisers physically cannot produce enough creative variants to let the model optimize. The AI creative tools don't just improve convenience — they are architecturally necessary for e-commerce to work at gaming-like conversion rates on this platform. Once creative volume catches up even partially, the model can do what it already does for gaming: test, learn, optimize across thousands of variants.
This reframes the entire e-commerce bull case. It's not "can AppLovin find e-commerce buyers?" — it's "can AppLovin close the creative volume gap so its existing model works on e-commerce the same way it works on gaming?" And the answer is: AI creative tools do exactly that, and they're 4 months away.
Foroughi laid out the unit economics of the platform with unusual precision:
The math Foroughi walked through:
My read: This is the clearest articulation of why APP's TAM isn't capped by gaming. The 1B+ daily users are already there. The ad attention is already there (30+ seconds per ad, ~50% are opt-in rewarded video). The model already works. The constraint is content diversity on the supply side. E-commerce + lead gen + other verticals solve that constraint. And the math is non-linear — it's not "add 10% e-commerce revenue" — it's "unlock 4x+ on the same impression base by improving conversion from 1.3% to 5%+."
I want to be careful here. Foroughi is the CEO making a pitch. But the logic is sound and the gaming proof-of-concept at >5% validates the ceiling is real.
Foroughi described the prediction problem in gaming: "We predict revenue all the way out to 28 days." The sequence: predict engagement with the ad → predict download → predict usage patterns → predict propensity to spend → predict amount of spend. All for a customer who didn't know they wanted the game. For companies running 20-30% margins.
"You can't get any of the sequence of predictions wrong to deliver the value for the advertiser. That's a really deep problem set. It took a lot of technology."
He then drew a parallel to LLMs: recommendation system models are "one of the best commercial uses of deep learning models" and share "a lot of the same architecture and technologies." AppLovin is developing with similar architecture AND paired with LLMs to accelerate development. Closing quote: "If we believe that AI technologies are going to be 2x more efficacious in 5 years... our system is going to be 2x more predictive... That would double our business or more."
My read: This is the moat narrative investors need. The competition concern (CloudX, Meta, LLM commoditization) assumes the prediction problem is shallow — one prediction, place an ad. Foroughi explained it's a deep sequence problem trained on 13+ years of data. CloudX can replicate the mediation layer. They cannot replicate the 28-day revenue prediction chain.
The product evolution over 18 months:
| Campaign Type | Launch | What It Does | Funnel Position |
|---|---|---|---|
| Universal | Mid-2024 | Mix of discovery + retargeting | Bottom (retargeting) + Mid |
| New Customer | Late Oct 2025 | Find users who visited but never purchased | Mid (prospecting) |
| New Visitor | ~Feb 25, 2026 | Find users who've NEVER seen the site | Top (pure discovery) |
Foroughi: "We started with something that was further down funnel. We moved to the middle of the funnel. Now last week, we rolled out new visitor campaigns."
On New Visitor: "That's the most powerful form of advertising. No one can debate how incremental something is if they find a customer they've never seen on their website before." Pilot results: "great results." Adoption: "really swift."
Key insight I missed before: The product roadmap is deliberately moving UP the funnel. They started with the easiest (retargeting — prove ROAS fast) and are systematically attacking more incremental, harder-to-prove campaign types. New Visitor is the hardest — pure discovery on a base of 1B+ gaming users — and they're already showing strong pilot results.
This is the same pattern as gaming. Start with performance (provable ROI), earn trust, then expand to higher-funnel. Except they're doing it in 18 months instead of 13 years because the model already exists.
Stumpf on the 20-30% guide: "There's definitely potential upside... when we initially said it a couple of years ago, we weren't even getting credit for 20%. It was just a baseline." They've grown "at a pace much greater than that."
The key framing: "This technology is very nascent. Our engineers are continuing to come up with directed model enhancements... as we grow and scale, we're getting more data to feed into it, which then improves the technology over time and continues to stack and compound."
My read: The recursive learning flywheel I've written about is being explicitly confirmed by the CFO. It's not just "we have a good model" — it's "the model gets better as we scale, which drives more scale." This is the core compounding mechanism. And he's saying the technology is "very nascent" — implying the improvement curve is still steep, not flattening.
| Team | Size | Notes |
|---|---|---|
| Total company | ~900 | |
| Core ad tech + back office | ~400 | |
| Gaming BD | ~100 | |
| E-commerce | ~15 | Incredibly lean |
| E-commerce expansion plan | "Tens of people" | Not hundreds |
Data center costs: growing at ~10% of revenue growth, and "beating that now." Model changes monitored for cost impact in real time.
Performance marketing: now running campaigns to acquire e-commerce advertisers onto the platform. Disciplined, same LTV/CAC framework. "We don't think that any of that should change the cost profile going forward."
Foroughi on why they don't ramp sales: "There's a whole bunch of platforms that ramped up sales, brought in new customers and hit a wall and never grew their business. And why did they hit that wall? Because the product itself was not good enough." They scaled gaming by showing ROI, not by begging for dollars. Doing the same with e-commerce.
My read: 15 people on e-commerce. Fifteen. And the product is showing "great results" and "swift adoption." This is the most operationally efficient growth expansion I've seen in tech. When GA launches and they add "tens of people," the incremental cost is noise vs the revenue potential.
Matt Cost asked about the "new player in mediation" (CloudX) without naming them. Foroughi's response:
My read: This was more comprehensive than anything on the earnings call. The key point: CloudX may compete on mediation. But mediation without the demand-side platform (the buying tools, the recommendation model, the 28-day prediction chain) is like building a stock exchange without market makers. The exchange itself is commodity infrastructure. The liquidity and matching engine are the moat.
Cost asked about AI disrupting game development. Foroughi's counter:
My read: This is the same counter-narrative from Q4, but more polished. AI doesn't commoditize gaming content — it makes good developers better, increases content volume, increases the need for a recommendation engine to sort through it. More content competing for the same eyeballs = AppLovin's matching becomes MORE valuable.
Stumpf on the recent interview about social media experiments: "It's less of like something that's core... it's really an opportunity for us to bring in new talent that's differentiated." Small R&D bets, won't change cost profile. Cross-pollination play.
My read: Non-event. Correctly filed under "exploration" not "strategy." Management isn't distracted.
Foroughi's final answer was the thesis in miniature:
| Topic | Prior (Q4 FY25 ER, Feb 22) | Post-Conference (Mar 4) |
|---|---|---|
| E-commerce product maturity | Self-service referral-only; GA "H1 2026" | Three campaign types live; full funnel (retarget → prospect → discover); more mature than I expected |
| Creative gap (43% breakage) | Red flag — unclear fix timeline | Now the key catalyst. 50x gap quantified. Multi-agent AI creative in pilot. Both static + video within 4 months. This is architecturally necessary, not nice-to-have |
| Gaming growth trajectory | 20-30% guide, unclear if conservative | CFO says "very nascent" technology + "definitely upside." It's a floor, not a ceiling |
| Conversion rate headroom | 1.3% current, >5% in gaming | New framing: content diversity unlocks 1.3% → 5%+ on same impression base. Non-linear TAM expansion. Potentially 4x+ gross platform value |
| Revenue model mechanics | Understood as platform fee | Now understand: spread on auction. Spread can WIDEN as conversion rate rises (not fixed share to publishers) |
| Competition (CloudX) | Expected manageable | Confirmed. Mediation without demand-side platform = stock exchange without market makers. 28-day prediction chain is unreplicable near-term |
| AI game disruption | Counter-narrative from Q4 | Now more polished. AI = more content = more need for recommendation engine. Moat strengthens |
| Operating leverage | Lean but unquantified | 15 people on e-commerce. 400 in core tech. Data center costs at <10% of revenue growth. Extreme leverage |
| 5-year thesis | E-commerce as TAM extension | Upgraded. Recommendation systems evolving like LLMs + AI 2x efficacy → system 2x predictive → doubles business. This is a compounding machine |
| Management quality | High credibility, track record clean | Higher. Foroughi's articulation of the 50x gap, the conversion rate math, and the 28-day prediction chain was CEO-as-chief-evangelist at its best. The "I named it AppLovin, that's a handicap" line was genuinely disarming |
This was the best investor presentation I've seen from AppLovin. Not because of new numbers — there were zero. Because Foroughi systematically dismantled every bear thesis with structural logic:
My thesis status moves from "Intact, on watch" to "Intact, strengthening." The conversion rate math alone — 1.3% today, 5%+ achievable with content diversity, on an impression base larger than SNAP+PINS+X+RDDT combined — is the single most compelling growth framework I've heard from any company this year.
My line in the sand hasn't changed: Q2 FY26 guide must show sequential acceleration AND e-commerce GA must have launched. But my confidence both will happen is now materially higher:
No position change. Holding 6.8% equity + 6.4% LEAPS (Jan'28 $450C). At $483, LEAPS are in the money. If Q1 actuals beat the sandbagged +6% QoQ guide — increasingly plausible — and creative tools ship on time, I'd add to equity on any pullback below $450.
The bear case hasn't changed (growth decelerates to 30%, re-rates to 12x = 40% downside). But the bull case is now better defined and more quantifiable than it was two weeks ago.
| Risk | Change | Note |
|---|---|---|
| E-commerce GA slips to H2 | Lower | 3 campaign types live; creative tools "within 4 months"; GA is a sales event not a product event |
| Gaming saturation | Lower | CFO: "very nascent" technology, "definitely upside" to 20-30% |
| Creative tools underdeliver | New, Medium | 50x gap is the key constraint. If AI creative doesn't close it, e-commerce conversion stays structurally below gaming |
| Q1 FY26 misses guide | Unchanged | Guide was conservative, New Visitor adoption is Q1 tailwind, but execution risk exists |
| CloudX gains share | Lower | Mediation ≠ demand platform. 28-day prediction chain, buying tools, publisher lock-in all intact |
| Conversion rate math is aspirational | New, Low-Medium | 5%+ is proven in gaming moments but platform-wide 5% requires massive content diversity + hundreds of thousands of customers. Multi-year path |
| Muddy Waters / short sellers | Unchanged | Not addressed. Market has moved on |
On gaming growth:
"There's definitely potential upside... this technology is very nascent." — Stumpf
On the creative gap:
"Game developers... they've got over 50,000 ads in a single campaign. The new category, e-commerce... at best, we saw 1,000 ads in a campaign." — Foroughi
On conversion rate:
"We today have a 1.3% conversion rate on our ads... We lose money on 98.7% of the ads we serve." — Foroughi
On the 5% ceiling:
"What happens when you get over 5%? It becomes one of the largest advertising platforms around." — Foroughi
On the 28-day prediction:
"Inside gaming, we predict revenue all the way out to 28 days... You can't get any of the sequence of predictions wrong." — Foroughi
On recommendation systems:
"Recommendation systems are going to evolve on the same trajectory as the large language models." — Foroughi
On operating leverage:
"We have something around 15 people for e-comm. So very, very small team." — Stumpf
On not ramping sales:
"We didn't scale gaming by begging for dollars, we scaled gaming by showing them that they make more money with their dollars spent on our platform than anywhere else in the world." — Foroughi
On AI + engineers:
"We've seen a 10x engineer become 100x in terms of output because of large language models that they pair with." — Foroughi
On self-awareness:
"I named the company AppLovin. That's a handicap. Nobody knows about the business." — Foroughi
Hold. Thesis strengthening materially. This was the most substantive investor presentation APP has given. The 50x creative gap, the 1.3% → 5% conversion math, and the 28-day prediction chain moat are now quantified and on the record. Next checkpoint: Q1 FY26 actuals (May 2026).
-wsm (Long APP equity 6.8%, Long APP Jan'28 $450C LEAPS 6.4%)
Sources: